430
Arşivlenen Tez
0
DOI Atanmış
0%
DOI Oranı
Arşivlenen Tez
Optimizing load balancing and task scheduling algorithms in cloud computing
The load balancing and task scheduling are important problems that need to be optimized in cloud computing in terms of meeting the expectations of both the user and the provider. A poorly optimized scheduling method harms the customer and the provider due to non-fulfillment of Quality of Service (QoS) and violation of the Service Level Agreement (SLA). It is possible to meet customer and provider demands in the best way with a well-optimized task scheduling algorithm. Metaheuristic algorithms produce good outcomes in solving NP-Hard problems such as cloud task scheduling problem. In this study, comparative performance analysis of metaheuristic algorithms such as Clonal Selection Algorithm (CSA), Genetic Algorithm (GA), Differential Evolution (DE), and Particle Swarm Optimization (PSO) are presented to optimize load balancing and task scheduling in cloud computing environments. Contrary to methods such as hybridization of metaheuristic algorithms and adaptive hyperparameters methods used in the literature to increase the search performance of algorithms, a parallelization method is proposed to increase search performance in this study. The proposed hybrid parallelization method is created by taking advantage of the global population master-slave and multiple-deme methods to be independent of the metaheuristic algorithm and to be implementing flexible. The proposed parallelization model is applied separately to CSA, DE, PSO, and GA algorithms, and detailed search performance analyses are presented. The results showed that when the normal sequential versions of the algorithms are compared, the CSA algorithm achieves more successful results than other algorithms, even though it gives close results to GA. The proposed parallelization model improves the performance of all tested algorithms, and the Parallel Clonal Selection Algorithm (PCSA) that parallel version of the CSA, reached better results than other compared algorithms.
Informed Monte Carlo tree search for board games
Developing artificial intelligence (AI) agents for adversarial game-playing using search-based methods presents the challenge of creating a robust utility function, which demands significant effort and specialized knowledge. Conversely, hastily devised simple utility functions often produce unsatisfactory outcomes. Monte Carlo Tree Search (MCTS) has emerged as a modern approach that avoids the need for such a strong utility function. Nevertheless, MCTS relies on a substantial number of game simulations to deliver accurate results, incurring notable computational expenses. This study introduces an inventive hybrid approach that leverages MCTS's strengths while seamlessly integrating a modified Upper Confidence Bound for Trees (UCB1) algorithm. This hybridization enhances MCTS's ability to exploit opportunities by including a basic utility function, reducing its reliance on a complex utility function. We conducted a series of experiments, applying this approach to classic board games like Tic-Tac-Toe, Mangala, and English Checkers. These experiments were compared to traditional Minimax and Alpha-Beta Pruning algorithms, along with the pure MCTS method.
Predicting sparse linear systems partition number using machinelearning
This thesis proposes a machine learning approach to solve sparse linear systems, a common problem in science and engineering. Our method uses machine learning to predict the optimal number of partitions needed for the block cimmino method. Traditional methods typically rely on trial and error or choosing the maximum number of cores as partitions to determine the best number of blocks. However, using a trained machine learning model eliminates this process. In this work, we present two models: one predicts two partition numbers with an AUC score of 89%, and the other predicts multiple partition numbers with an AUC score of 76% for the block cimmino method. We achieve this by using previously identified features in the literature and applying them in a novel context. We trained and tested our models using a diverse set of matrices, incorporating feature selection, demonstrating the effectiveness of leveraging established features in new applications and the robustness of our model in addressing the challenges of partitioning sparse linear systems. Our proposed method is faster and more effective than traditional methods.
Machine learning-based feature selection approach for no-show rate prediction: A case of aviation industry
In today's environment of increasing competition, rapidly changing customer demands, and globalization, companies are compelled to operate with low profit margins and adapt swiftly to changing conditions to gain a competitive advantage. This thesis was conducted during a period when the importance of technologies such as artificial intelligence and big data is increasing in the highly competitive aviation industry. The aviation industry is known for its high capital requirements, low profit margins, intense competition, complex operational structures, and constant pressure from local or global conditions. Considering the low profit margins in the aviation industry, airlines are continually encouraged to enhance their efficiency, reduce operational costs while maintaining customer satisfaction, or explore areas that could generate additional revenue. A significant proportion of passengers who purchase tickets do not show up for their scheduled flights. This situation allows airlines to practice overbooking, and accurately predicting the rate of no-shows can provide substantial additional revenue. However, if all passengers do show up, airlines would incur additional costs and suffer reputational damage. Therefore, accurately predicting no-show rates and optimizing operational planning processes are critically important for airlines. No-show rates are generally estimated by analyzing historical data using statistical methods. However, recently developed machine learning and big data analytics have the potential to play a significant role in solving such complex problems. There are very few studies in the literature on predicting no-show rates using machine learning methods. These studies typically use only Passenger Name Record (PNR) data for no-show prediction, neglecting many external factors. Using only passenger information reduces the accuracy of predictions due to the unique characteristics of each passenger. Unlike other studies, this research investigates the impact of external factors such as weather conditions, traffic density, special occasions, and public holidays on no-show rates without using PNR data, employing big data analysis and machine learning methods. This study focuses on the unique and critical need to improve operational planning processes in the aviation industry. It aims to contribute to understanding and addressing the challenges in the aviation industry, particularly in accurately calculating the no-show rate. To achieve this goal, the importance and effects of accurately predicting the no-show rate have been comprehensively examined, and a model has been proposed using machine learning algorithms to address the problem at hand.
Malware detection using machine learning and evolutionary algorithms
As cyber threats grow in complexity and frequency, the need for robust and adaptive malware detection mechanisms is critical due to their impact on national security and economic stability. Given malware's evolving nature to evade detection, selecting effective parameters by examining the interactions between malware characteristics is crucial. Traditional signature-based systems are becoming inadequate as malware adapts to new technologies. To address this, a new system is developed in this thesis that focuses on malware behavior and feature relationships. Multi-objective genetic algorithms (MOGAs) are employed to identify critical features for detection, which are then utilized by machine learning (ML) algorithms within a hybrid framework to accurately detect and classify malware. The objective of this thesis is to determine the optimal feature selection and classification methods that yield the highest accuracy within the Cuckoo Sandbox environment. Specifically, classifiers such as the J48 Decision Tree (J48), Reduced Error Pruning Tree (REP Tree), Adaptive Boosting Model 1 (AdaboostM1), Multilayer Perceptron (MLP), and Naive Bayes (NB) were evaluated. As a result of the analysis, the feature set was reduced from 335 to 200 by considering the relationships between features, resulting in a high accuracy of 93.33% and a performance improvement of 40% due to the reduction in the number of features. The obtained metrics were meticulously compared and evaluated with respect to the employed algorithms and methodologies. Furthermore, Mc Nemar's test was utilized to assess the performance of different malware detection classifiers by comparing their correct and incorrect classifications. The results from Mc Nemar's test indicated significant improvements, highlighting the effectiveness of the proposed system.
Assessment of natural lighting levels in studio classrooms of architectural departments using the dialux EVO12 tool
Natural light plays a crucial role in shaping the ambiance and functionality of interior spaces in all types of buildings. The architectural studio buildings are where students practice their daily architectural design activities. Lighting is more important for work and comfort than in regular educational classrooms because of the time taken and the particular focus this activity requires. This research discusses the effects of natural light on architectural studios in AYBU, Amasya, and Samsun universities. The study examines how natural light achieves users' comfort levels in the three university buildings. The method uses the DIALux tool to model the building in a 3D model and run a simulation on working days during the educational year. This simulation aims to obtain the levels of lighting and illumination and compare them with the design standards between universities. In addition, the research explores the improvements that can reduce the amount of light entering rooms, namely solar refractors. The results showed that natural lighting levels at Amasya University are very high in architectural studios, which causes glare and thus hinders the educational process. This is due to the location of the rooms in the east of the building and the large size of the windows. The AYBU, a laboratory building, has been reused as an educational building for the Department of Architecture. Light levels are higher in the summer and lower in the winter, but can be remedied. Samsun University, a tobacco factory, was converted into an educational building, and part of it was allocated to the Department of Architecture. It was the best among the three universities in terms of lighting. Studios only suffer from the lack of light entering rooms at some times of the year, mainly in the winter. Solar breakers have been proposed as a solution to improve lighting levels. It is because Samsun University used it, but it is fixed, and this causes a lack of lighting in the winter. The test implemented for Amasya University and AYBU has yielded acceptable results. It gives us an indication of the possibility of using it as an architectural solution, but you need to specify the specifications of the solar refractors for each university in terms of shape, size, and manufacturing material, and it is preferable to use them mobile with a solar light sensor.
Behavior of shear deficient reinforced concrete columns jacketed with engineering cementitious composite concrete for seismic loads
This study focuses on treating the lack of shear resistance of columns exposed to axial and lateral force, such as earthquake and wind loads. The original columns were strengthened using Engineered cementitious composite ECC jackets, which are characterized by high plastic energy dissipation capability, and a mechanism that allows for the development of multiple tensile microcracks, high tensile resistance, and high hardening strain in tension, using two types of ECC, coated and uncoated fibers. This study bases on simulation and finite element analysis of 40 retrofitted models of experimental columns using the ABAQUS software under conditions similar to those that occurred in the laboratory, by utilizing concrete damage plasticity model to represent concrete behavior under different conditions. The stress-strain relationship and constitutive models in compression and tension have been calibrated. The finite element verification was implemented for all experimental specimens, which exhibit high identical behavior with the tested columns. All models of strengthened specimens with Engineered cementitious composite (ECC) exhibit an increase in lateral load capacity as well as a distinct increasing in ductility that change the sudden shear failure to flexural failure and high enhancement in total and plastic dissipation energy compared with the original experimental columns. The ECC demonstrate excellent compatibility with steel reinforcement and large ability to compensate for the deficiency of shear reinforcement.
Analysis of stability-deformation properties of a cut slopes under seismic effects
This thesis investigates the seismic slope stability of a critical cut slope segment (Km 13+580) along the Palu-Beyhan-Gökdere Highway in Eastern Anatolia, Turkey, a region known for high seismic activity. Utilizing finite element analysis with PLAXIS 2D software to model the slope geometry and soil properties. The Hardening Soil small-strain (HS small) model is utilized to accurately capture the nonlinear soil behavior, particularly under dynamic loading. Appropriate boundary conditions and ground motion data from the PEER ground motion database are incorporated into the analysis. The study evaluates the slope's stability and displacement under seismic loading. The data for field studies conducted by an existing project, including geotechnical surveys and soil sampling, were used to inform the model parameters. Static and dynamic analyses are employed to understand failure mechanisms. The static analysis assesses current slope stability, while the dynamic analysis identifies critical areas susceptible to failure during earthquakes. Findings show the Km 13+580 slope is susceptible to significant displacements and stress concentrations under earthquake loads, potentially leading to slope failure. Based on these results, the study proposes mitigation measures to reduce the risk of failure, ensuring highway safety and reliability. This includes enhancing the existing successful soil nail design, demonstrably reducing deformation compared to the previous condition.
Removing metal ions from the solution of electropolished kovar alloy
Electropolishing is a metal surface treatment technique, that involves the selective removal of particles from the metal's outer surface via the application of a controlled potential in an ionic solution. The aim of electropolishing is to obtain a smooth surface by removing ions on the metal surface in an appropriate solution to provide better performance parameters and improve the resistance of the parts against corrosion. However, the dissolution of metal ions during electropolishing leads to significant solution contamination, necessitating frequent replacement. This study focuses on collecting metal ions generated during Kovar alloy electropolishing onto a copper foil through electrodeposition, thus enabling the reuse of the electropolishing solution. Electrodeposition experiments were conducted using graphite or titanium anodes and copper foil cathodes. The optimal results were obtained with a voltage of 13 applied for 2 h, maintaining a pH of 3 at 25°C. This approach efficiently cleanses electropolishing solutions, fosters sustainable use and minimizing environmental impact.
Environmental sound database design and implementation
This thesis proposes "An Environmental Sound Database Design and Implementation" for environmental sound recognition. Environmental sound recognition has become a hot topic in recent years. Environmental sound recognition can be used in security systems, in robot navigation, in internet search engines and more. It is necessary to establish a database which comprises the sounds to be recognized in advance to be able to recognize the environmental sounds. Sound database development is the first stage of all environmental sound recognition tasks. In this study, a database based relational database system is designed.. The sound databases designed and used so far are all constructed on the directory and file system of the operating system. For the first time, a sound database system was constructed using relational database and SQL (Structured Query Language) was used. After database design, the example sounds was loaded, a GUI (Graphical User Interface) as a desktop application and a web interface for accessing over internet was designed.